Handwritten Customer Feedback Form Mining in the FMCG Sector Using Meta-Learning

For fast moving consumer goods (FMCG) companies, customer feedback is a very important component, yet a substantial portion of this feedback is still collected in handwritten form through surveys, complaint registers, retail feedback cards, and field audits. The diversity of handwriting styles, scripts, languages, and noisy acquisition conditions poses significant challenges for large-scale automated analysis. These challenges are further amplified by the multimodal nature of feedback forms, which often contain tables, numerical entries, symbols, and company logos, making information extraction complex even from scanned printed templates. This paper proposes a meta learning-based model for handwritten customer feedback analysis in low-resource and real-world FMCG environments. The proposed system integrates form layout detection using YOLOv8 (You Only Look Once Version 8), handwritten text recognition using CRNN (Convolutional Recurrent Neural Network)/ TrOCR (Transformer Based Optical Character Recognition) and sentiment analysis using BERT (Bidirectional Encoder Representation from Transformers) within a unified pipeline. Meta learning is incorporated as a training strategy to enable rapid adaptation to new document layout and unseen handwritten styles with minimal labelled data. Unlike conventional that require large datasets, the proposed model can efficiently adapt to diverse tasks using few shot learning, this significantly improve robustness against domain variability, noisy input and limited data availability. Experimental results demonstrate that the proposed framework achieves reliable performance in layout detection, text recognition and sentiment analysis while reducing the dependency on large labelled datasets. The proposed approach supports near real-time analysis to facilitate faster decision-making and bridge the gap between handwritten text recognition research and FMCG analytics.

Authors

Institutions

Publication Details

Journal
Jharkhand Journal of Development and Management Studies
Published
2026-11-18
DOI
https://doi.org/10.70994/jjdms.11463.11480
Primary Topic
Handwritten Text Recognition Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Handwritten Customer Feedback Form Mining in the FMCG Sector Using Meta-Learning

Goutam Sanyal, Madhumita Singha
1 links
Jharkhand Journal of Development and Management Studies
Handwritten Text Recognition Techniques
article

Handwritten Customer Feedback Form Mining in the FMCG Sector Using Meta-Learning

Goutam Sanyal, Madhumita Singha
article en

Abstract

For fast moving consumer goods (FMCG) companies, customer feedback is a very important component, yet a substantial portion of this feedback is still collected in handwritten form through surveys, complaint registers, retail feedback cards, and field audits. The diversity of handwriting styles, scripts, languages, and noisy acquisition conditions poses significant challenges for large-scale automated analysis. These challenges are further amplified by the multimodal nature of feedback forms, which often contain tables, numerical entries, symbols, and company logos, making information extraction complex even from scanned printed templates. This paper proposes a meta learning-based model for handwritten customer feedback analysis in low-resource and real-world FMCG environments. The proposed system integrates form layout detection using YOLOv8 (You Only Look Once Version 8), handwritten text recognition using CRNN (Convolutional Recurrent Neural Network)/ TrOCR (Transformer Based Optical Character Recognition) and sentiment analysis using BERT (Bidirectional Encoder Representation from Transformers) within a unified pipeline. Meta learning is incorporated as a training strategy to enable rapid adaptation to new document layout and unseen handwritten styles with minimal labelled data. Unlike conventional that require large datasets, the proposed model can efficiently adapt to diverse tasks using few shot learning, this significantly improve robustness against domain variability, noisy input and limited data availability. Experimental results demonstrate that the proposed framework achieves reliable performance in layout detection, text recognition and sentiment analysis while reducing the dependency on large labelled datasets. The proposed approach supports near real-time analysis to facilitate faster decision-making and bridge the gap between handwritten text recognition research and FMCG analytics.

Jharkhand Journal of Development and Management StudiesVol. 24(1-4)
Xavier Institute of Social Service (IN), Raiganj Surendranath Mahavidyalaya (IN)
Handwritten Text Recognition Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon